Office Letter Entity Annotation

From BioIE Wiki

Main Page  : Entity  : Office Letters  : Office Letter Entity Annotation

Chaining

Around December 2003, we developed and introduced a chaining tool that lets us annotate discontinuous entities in order to make connections explicit in the actual annotation by building a chain of strings and applying a single tag to them. In office letter annotation, however, we do not use chaining.

Modifiers

Instead of chaining, annotators include modifiers in a single tag as long as they are in the same noun phrase. For more information, see Noun Phrases in the Sign/Symptom Decision Tree.

Examples

Q: Given the term "soft 1/6 vibratory systolic ejection murmur", I'm just tagging "systolic ejection murmur" as S/S and leaving the rest as Status (per Dr. Mandel's instructions from his e-mail on 12/13). Is this still correct? And this brings us back to the question of how we're dealing with adjectives. I'm not sure I'm being entirely consistent.

A: Include the modifiers as in the example above.

Q: Again on the topic of modifiers, given "exercise-induced ectopy", should we include "exercise-induced"? How about "complex" in "complex arrhythmia", "strong" in "strong palpitations" and "neurally-mediated" in "neurally-mediated syncope"?

A: Yes, include them.

Q: What about "notched T waves" and "prominent U waves"? What should we do when we come across a word or term that describes the heart beat, but which, to the layman, doesn't clearly denote an abnormality?

A: "T waves" and "U waves" by themselves refer to parts of the EKG and would be considered Vital Signs. However, "notched T waves" and "prominent U waves" indicate abnormalities and should be tagged as Sign/Symptom. The same for any similar phrase.

Unless the adjective is something like "normal" or "unremarkable"; those, in a later run, we would tag as follows.

Annotating Discontinuous Phrases

Q: Are we chaining? Specifically, in the case of "low aerobic and physical working capacity", can I get "low aerobic capacity" and "low physical working capacity"?

A: No, we are not chaining. So in this case you will have to tag the whole string "low aerobic and physical working capacity". And tag it as Sign/Symptom, since it is pathological. Working through the Sign/Symptom Decision Tree:

Here we got into discussion.

• This phrase is quite common in these letters, even though it combines two observations each of which would be a Sign/Symptom by itself ("low aerobic working capacity", "low physical working capacity").
• And it's a single noun phrase (next paragraph).
• Therefore, TAG THE WHOLE PHRASE AS A SINGLE SIGN/SYMPTOM:

It would be different if the modifier "low" were separated from the things it modifies, as in "aerobic and physical working capacity were both low". Then we would have to tag "low" as Status and "aerobic and physical working capacity", in the absence of chaining, as a single Vital Sign.

Syntax Issues Within String

Q: I know we would tag "prolonged QT interval" as a S/S, but what if the syntax doesn't work out in our favor, such as "Her QTc was noted to be 495 msec, which is prolonged." Can we salvage that somehow?

A: Not at this stage, while we are tagging only Sign/Symptom and Condition. In complete tagging, this should become

Semantics vs. Syntax

Q: Can we tag the same medical problem sometimes as S/S and sometimes as Condition depending on context? For example, I've been tagging tachycardia and syncope as S/S, but in one particular letter, the doctor says that his patient is "followed with the diagnosis of congenital long QT syndrome, ventricular tachycardia, and neurally-mediated syncope," seemingly regarding all three as preexisting conditions. I guess I just feel weird tagging long QT as a condition and the others as S/S when they appear together in a list suggesting that they fall into the same category.

A: There are three issues here: context, the span of the references tagged, and the tags.

• Context: The wording of diagnoses in these office letters is largely controlled by the standard codes required by insurance companies. Their idea of "diagnosis" isn't exactly the same as ours, and includes Sign/Symptom alongside Condition. So don't go by the doctor's sentence structure and application of the word "diagnosis"; feel free to use your own judgment. You're not disagreeing with the doctor, you're just using a different classification than the insurance company. As Yang put it on the onco-sr list,

'We always tag what the author really means, rather than tag the mentions literally. However, in this example, I would tag the first (long QT syndrome) as Disease/Condition, and the rest as S/S. The doctor just didn't do a good job paralleling these phrases up, and I believe he didn't mean "long QT syndrome" as symptom.'

• Span: We ARE going to include the clinically relevant modifiers "ventricular" and "neurally-mediated".

• Tags: As you said, Condition, Sign/Symptom, Sign/Symptom, respectively.

So the tagging of this phrase would be as follows.

Tag Within Tag

Q: The letter states that the patient has a history of "transitional AV canal status post repair" and underwent surgery for "closure of a residual mitral valve cleft." The previous annotator has tagged the entities in quotes as S/S. The underlying problems are presumably "transitional AV canal defect", which isn't stated in full, and "residual mitral valve cleft", which is. How should these be tagged? Should these entities be saved to be tagged under some sort of "medical procedure" tag later? Are we doing tags-within-tags? Specifically, even if "closure of a residual mitral valve cleft" might later be tagged as "procedure", should I still tag "residual mitral valve cleft" now?"

A: Both of those strings will be tagged as Treatment Action. We will be doing tag-within-tag, so that should not prevent you from tagging "residual mitral valve cleft" as Condition (if that's what it is).

Tagging Masked Text

Q: How do we tag the following sentence?

I am not sure if the word or words blocked out by the tag <NAMES_1> apply to the murmur. Whether or not the tag <NAMES_1> is applicable to the murmur, how should I tag this particular term?

A: Treat "<NAMES_1>" as part of the text and follow the modifiers rule. We can tag this phrase as a [S/S]. See below.

In general, if it seems that the anonymizer has mistakenly masked part of a taggable term,

• Check WhoNamedIt.com. It's a great resource to check for names that should not be masked.

• Tag the text as though it weren't masked.

• Notify the list under subject: "confidentiality", including your decision to tag it.

Main Page  : Entity  : Office Letters  : Office Letter Entity Annotation